JP5298393B2 - Parallel Reed-Solomon RAID (RS-RAID) architecture, device, and method - Google Patents

Abstract

The parallel RS-RAID data storage architecture can aggregate that data and checksums within each cluster into intermediate or partial sums that are transferred or distributed to other clusters. The use of intermediate data symbols, intermediate checksum symbols, cluster configuration information on the assignment of data storage devices to clusters and the operational status of data storage devices, and the like, can reduce the computational burden and latency for the error correction calculations while increasing the scalability and throughput of the parallel RS-RAID distributed data storage architecture.

Description

  This application claims the benefit of US Provisional Application No. 60 / 989,670 “Parallel RAID Implementation for RAID 6 and Reed-Solomon Code” filed on Nov. 21, 2007, which is hereby incorporated by reference in its entirety. Includes all cited references incorporated into.

  The RAID (Redundant Array of Inexpensive Disks) architecture uses a group of data storage units such as hard disks to provide a fault-tolerant data storage device. The RAID architecture uses a forward error correction (FEC) code and a spare data storage unit to protect information from errors and disk failures. An information symbol may be a bit, byte, or word. Information symbols may be encoded to form code symbols including data and checksum or parity symbols. In the case of a systematic forward error correction code, the information symbol can be explicitly represented in the data symbol part of the code symbol.

  The Reed-Solomon code can be used in a RAID architecture (RS-RAID) to allow a number of storage unit failures equal to the number of checksum symbols. For example, a quad error correction RS-RAID architecture that allocates 20 storage units for data and 4 storage units for checksum can tolerate failures in up to four storage devices, including four.

  RS-RAID architecture typically uses a single RAID controller to protect data symbols written to the data storage unit. When a single RAID controller is used to perform checksum, encoding, and decoding calculations, the RAID architecture's throughput or data storage and retrieval speed is not RAID and data storage is not fault tolerant. May be reduced compared to architecture.

  Thus, a distributed data storage architecture with high throughput and fault tolerance may be desirable.

  In a high performance storage architecture, multiple RAID controllers may communicate with each other through a common set of communication paths called a communication fabric. The communication fabric may have a high latency compared to the communication path between the RAID controller and the storage device assigned to a given RAID controller. A high-latency communication fabric can reduce the throughput of a RAID data storage architecture if the traffic between RAID controllers, such as data, messages, and configuration traffic, does not match the fault-tolerant distributed data storage task. . Each RAID controller, which may intervene between the communications fabric and the assigned collection of data storage devices, may be referred to as a data storage architecture node. The RAID controller and assigned data storage device may be referred to as a data storage cluster.

The Reed-Solomon RAID (RS-RAID) architecture can protect redundant information storage devices to protect information symbols written to and read from storage devices such as hard disks . An RS-RAID architecture that uses m checksum devices can tolerate as many as m simultaneous failures of data storage devices. The checksum symbols of _m may be denoted by c ₁ , c ₂ ,..., _cm . RS-RAID architecture _{also, d 1, d 2, ...} , it may include a data storage device of the number n of information holding or data symbols indicated by _{d n.}

  The checksum and data storage device may store data and checksum symbols as bits, bytes, words, etc. It may be noted that certain types of forward error correction codes (FEC), such as Reed-Solomon (RS) codes, typically use bytes. For example, the RS code may operate on a byte block, such as a block that encodes 233 information bytes into 233 data bytes and 32 checksum bytes in a 255 byte block.

RS-RAID architecture, the corresponding data storage device _{D 1,} D 2, ..., the data symbols _{d 1,} d 2, which is held by the _{D n,} ..., using the _{d n,} i-th checksum device _{C i} The checksum symbol c _i stored in can be calculated. RS-RAID _{architecture, D 1, D 2, ...} , D n, C 1, C 2, ..., if any of the following storage devices m of the storage device _{C m} fails, among failed device Each c _i (1 ≦ i ≦ m) may be determined so that the content of any of the devices can be reconstructed from intact or non-failed devices. The RS-RAID architecture can provide fault tolerant operations due to the characteristics of the van der monde matrix, which calculates and maintains checksum symbols, and the data and checksum symbols read from the storage device. Used to recover information from. The RS-RAID controller calculates the data in the storage device by calculating the inverse matrix of the (n × n) portion of the added and expanded van der Monde matrix and unit matrix even if the storage device fails. The checksum symbol may be recovered.

換言すれば、データおよびチェックサムシンボルが、それぞれ、（ｎ×１）次元および（ｍ×１）次元ベクトルＤ＝［ｄ_１，ｄ_２、…、ｄ_ｎ］^ＴおよびＣ＝［ｃ_１、ｃ_２、…、ｃ_ｍ］^Ｔとして表され、また、関数Ｆ_ｉが行列Ｆの行として表される場合、ＲＳ−ＲＡＩＤアーキテクチャは、チェックサムシンボルを式２ａによって符合化しうる。
Ｃ＝ＦＤ （式２ａ）
式２ａは、

に等しい。 To generate the checksum symbol, the RS-RAID architecture may weight the data symbols with the elements of the van der Monde matrix and sum the weighted data symbols using the linear function F _{i according} to Equation 1. Since the function F _i can be obtained from the i th row of the elements of the van der Monde matrix, F _i = [f _{i, 1} ; f _{i, 2} ;... F _{i, n} ] ^T.

In other words, the data and checksum symbols are represented by (n × 1) dimensional and (m × 1) dimensional vectors D = [d ₁ , d ₂ ,..., D _n ] ^T and C = [c ₁ , c, respectively. _{2, ...,} expressed as c ^{m] T,} also, if the function _{F i} is represented as rows of the matrix F, RS-RAID architecture, may be encoded by the equation 2a checksum symbols.
C = FD (Formula 2a)
Equation 2a is

be equivalent to.

として書かれうる。 In the case of an advantageously designed RS-RAID FEC code, the F matrix may be an (m × n) van der Monde matrix with elements: f _{i, j} = j ⁱ⁻¹ . Where the indices i = 1, 2,..., M and j = 1, 2,..., N correspond to the rows and columns of the van der Monde matrix, respectively, and algebraic operations use the properties of Galois fields. Implemented. For example, the (3x4) van der Monde matrix is

Can be written as

を回復するために、並列ＲＳ−ＲＡＩＤアーキテクチャは、ファンデルモンデ行列およびＩで示す添加された（（ｎ−ｍ）×（ｎ−ｍ））単位行列を含む拡大されるかまたは分割された行列Ａの逆行列を求め、Ａの逆行列に、作動可能な記憶ユニットから読出されるデータおよびチェックサムシンボルの集合ＤおよびＣをそれぞれ右から乗算し(ｐｏｓｔ−ｍｕｌｔｉｐｌｙ)うる。記号的には、回復される情報ベクトル

は、

から得られうる。式中、拡大行列は

であり、

は、拡大されたデータおよびチェックサムシンボルベクトルである。表記Ｉｎｖ（Ａ）は、正則な（ｎ×ｎ）正方行列を形成するＡの行の部分集合の逆行列などのＡに基づく逆行列をもたらし、また、以下で述べるように、

で示す列行列Ｅのｎの行の、対応する選択されるかまたは選抜された集合に共形的である関数であると理解されてもよい。Ａ行列の逆行列を求めるプロセスは、Ａの行の選択された集合の反転とみなされてもよく、選択は、作動可能なデータ記憶デバイスのリストおよび行列にベクトルを掛ける計算における共形性についての要件によって確定される。（（ｎ＋ｍ）×ｎ）拡大行列Ａのｎの行の全ての部分集合は、Ｆがファンデルモンデ行列であるため反転可能であることが留意されてもよい。 (N × 1) information vector from a codeword or code symbol that may contain errors

In order to recover the parallel RS-RAID architecture, an expanded or partitioned matrix including a van der Monde matrix and an added ((n−m) × (n−m)) identity matrix denoted I An inverse matrix of A can be determined, and the inverse matrix of A can be post-multiplied from the right, respectively, by the data read from the operable storage unit and the sets D and C of checksum symbols. Symbolically, the recovered information vector

Is

Can be obtained from Where the expansion matrix is

And

Are the expanded data and checksum symbol vectors. The notation Inv (A) yields an inverse matrix based on A, such as the inverse of a subset of the rows of A that form a regular (n × n) square matrix, and, as described below,

May be understood to be a function that is conformal to the corresponding selected or selected set of n rows of the column matrix E. The process of finding the inverse of the A matrix may be viewed as an inversion of the selected set of rows of A, and the selection is for conformality in the computation of multiplying the list and matrix of operable data storage devices by a vector. Determined by the requirements of It may be noted that all subsets of the n rows of the ((n + m) × n) augmented matrix A can be inverted because F is a van der Monde matrix.

は、

として表されうる。式中、共形性は、行列Ａの選択された部分を反転する前に、ＥおよびＡの対応する行を選択することによって実施される。 In expanded form, the expression

Is

Can be expressed as Where conformality is implemented by selecting the corresponding rows of E and A before inverting the selected portion of matrix A.

が形成されうる。換言すれば、Ｉｎｖ（Ａ）＝（Ａ'）^−１であり、また、

である。たとえば、４＋２ＲＳ−ＲＡＩＤアーキテクチャの場合、回復されるかまたは復号化されるデータ

は、

によって、拡大されたファンデルモンデ行列の最初の４つの行および記憶デバイスアレイから読出されたデータおよびチェックサムの最初の４つのエンティティから抽出された回復されるデータシンボルのベクトルでありうる。 In other words, the respective storage devices of RS-RAID architecture, the row of the augmented matrix A and a column vector _{E = [d 1, d 2} , ..., d n, c 1, c 2, ..., c m] corresponding ^T Can be represented by If no redundant storage device in m redundant storage devices fails, the recovered information symbols are determined by selecting any subset of A's n rows and E's n corresponding elements. A square matrix A ′, which may be described as a data recovery matrix, and a vector of data read from the corresponding data storage unit

Can be formed. In other words, Inv (A) = (A ′) ⁻¹ , and

It is. For example, in the case of a 4 + 2RS-RAID architecture, the data to be recovered or decoded

Is

Can be a vector of recovered data symbols extracted from the first four rows of the expanded van der Monde matrix and the data read from the storage device array and the first four entities of the checksum.

は、以下の通りに、作動可能なデバイスに対応する４つの行を選択することによって、Ｅ'から回復されうる。

式中、２重末梢線は記憶デバイスの故障を示し、１重末梢線は、逆行列を形成し、その後の計算を実施するための記憶デバイスの選択解除を示しうる。逆行列は、ガウス消去法または別の方法によって計算されてもよい。

の値が得られると、回復されるかまたは推定される任意のチェックサムベクトル

が、

を使用して、データベクトル

に基づいて計算されてもよい。 For example, if the third, fifth, or both third and fifth data storage devices fail,

Can be recovered from E ′ by selecting the four rows corresponding to the ready devices as follows.

Where the double peripheral line indicates a failure of the storage device and the single peripheral line forms an inverse matrix and may indicate deselection of the storage device for performing subsequent calculations. The inverse matrix may be calculated by Gaussian elimination or another method.

An arbitrary checksum vector that is recovered or estimated once the value of is obtained

But,

Use the data vector

May be calculated based on

  A parallel RS-RAID data storage architecture may aggregate data and checksums within each cluster into intermediate or partial sums that are transferred or distributed to all clusters. Use of intermediate data symbols, intermediate checksum symbols, cluster configuration information regarding allocation of data storage devices to clusters, and status regarding operation of data storage devices, etc., improves scalability and throughput of parallel RS-RAID distributed data storage architecture. However, the calculation load and latency for error correction calculation can be reduced.

  The present disclosure will refer to the accompanying drawings in which like numerals represent like elements.

It is a figure which shows the example of parallel RS-RAID (Reed-Solomon redundant array of inexpensive disk). It is a figure which shows the example of a structure matrix. It is a figure which shows the example of a RAID controller. It is a flowchart of the example of a checksum program. It is a flowchart of the example of a checksum update program. It is a flowchart of the example of a data program.

  FIG. 1 is an example of a parallel RS-RAID architecture 100 for data storage. The parallel RS-RAID architecture 100 may include a communication fabric 1200, RAID controllers 1111 to 1113, and storage devices 1001 to 1012. Storage devices 1001-1004, 1005-1008, and 1009-1012 may be coupled to RAID controllers 1111-1113, respectively. In other words, a subset or cluster of storage devices 1001-1012 can be coupled to their corresponding RAID controllers 1111-1113. The number of storage devices that can be coupled to each RAID controller 1111 to 1113 may or may not be equal, and the storage devices are configured for the RAID controller to improve fault tolerance, improve throughput, and the like. Or the mapping may change dynamically. For example, the assignment of storage devices 1001-1012 to RAID controllers 1111-1113 may be determined by a configuration matrix or similar data structure.

  FIG. 2 shows an example of a configuration matrix 200 that may include a variable “t”, where the variable “t” is an index or counter of the number of RAID controllers. For example, the row 206 of the configuration matrix 200 indicates the mapping function Q (t) between the RAID controller index number “t” and each RAID controller such as the RAID controllers 1111 to 1113. Row 202 shows a RAID storage device start index QS (t), and row 204 shows a RAID storage device end index QE (t). For example, QS (2) = 1005 and QE (2) = 1008. Since the device number offset may be supplied by the function J (), J (QS (2)) = 5, for example, the fifth storage device starts with the second storage device cluster. It may be noted that this can be shown. The configuration matrix 200 may map storage devices to corresponding RAID controllers. In other words, the configuration matrix 200 may determine or control which subset or cluster of storage devices are assigned to a given RAID controller. For computation, the configuration matrix 200 updates or maintains the checksum and data, which can determine the beginning and end of a weighted partial sum that can be encoded or decoded, as described below. It can be done.

  Communication fabric 1200 may couple input and output (I / O) digital signals between RAID controllers 1111-1113 and between parallel RS-RAID architecture 100 and external devices. For example, the communication fabric 1200 may combine digital signals such as data symbols, checksum symbols, intermediate data, checksum symbols, etc. between RAID controllers 1111-1113. The communication fabric 1200 may use a parallel bus structure, a serial data link, an optical backplane, or the like. The communication fabric 1200 may use one type of bus, link, or backplane structure for external communication and another type for communication between the RAID controllers 1111-1113.

  The RAID controllers 1111 to 1113 may calculate data checksum symbols for each storage device in the allocated subset of storage devices or clusters given by a configuration matrix such as the configuration matrix 200 or other data structure. The RAID controllers 1111 to 1113 totalize or accumulate partial sums of error correction code calculations, and the aggregated data and parity calculation results are transmitted to other RAID controllers in the parallel RS-RAID architecture 100 through the communication fabric 1200. Can be reported. Details of the partial sum calculation for data and checksum symbols may be described with reference to a particular RAID controller, but the corresponding calculation can be performed in any parallel RS-RAID architecture 100, such as RAID controller 1111. It may be implemented by a RAID controller.

  FIG. 3 shows an example of a RAID controller 1111 that may include a communication fabric interface 1111a, a RAID control unit 1111b, an intermediate sum device 1111c, a storage device interface 1111g, and a storage device failure sense unit 1111h. Communication fabric interface 1111a may couple signals to and from a communication fabric such as communication fabric 1200 to intermediate sum device 1111c and RAID control unit 1111b. The RAID control unit 1111b may be coupled to the intermediate sum device 1111c, the storage device interface 1111g, and the storage device failure sense unit 1111h. The storage device interface 1111g may be coupled to the RAID control unit 1111b, the intermediate sum device 1111c, and the storage device failure sense unit 1111h. The RAID controller 1111 can be coupled to and from the communication fabric 1200 as described above, and can be coupled to and from the storage devices such as the storage devices 1001 to 1004 via the storage device interface 1111g.

  The intermediate sum device 1111c may include an intermediate sum calculator 1111d, a recalculator 1111e, and a calculation control 1111f. The intermediate sum calculator 1111d may be coupled to the communication fabric interface 1111a, the storage device interface 1111g, the recalculator 1111e, and the calculation control 1111f. The recalculator 1111e may be coupled to the communication fabric interface 1111a, the intermediate sum calculator 1111d, the calculation control 1111f, and the storage device interface 1111g. Calculation control 1111f may couple to intermediate sum calculator 1111d, recalculator 1111e, and storage device interface 1111g.

  The communication fabric interface 1111a transfers an information symbol between the parallel RS-RAID architecture 100 and an external device, and an information symbol, a predetermined part of the information symbol, a data symbol between the communication fabric 1200 and the RAID controller 1111 Checksum symbols such as intermediate checksum symbols, control signals, clock signals, and the like. The communication fabric interface 1111a can reformat information symbols from bits to bytes, words, or other symbols, can multiplex and demultiplex signals, can synchronize data transfer, and buffers signals by line drivers and receivers You can ring and so on. In other words, the communication fabric interface 1111a may adjust digital signals for transmission over a communication fabric such as a digital bus, may buffer data transfers, and so on.

  RAID control unit 1111b may receive signals from communication fabric interface 1111a and storage device, may select data symbols from a subset of information symbols, may strip data and checksum symbols across storage devices, and forward error correction code The operation of the intermediate sum device 1111c can be controlled by (forward-error correction code) (FECcode). For example, the subset of information symbols may be information symbols that are represented by data symbols and stored in an operable data storage device that is controlled by RAID controller 1111. The intermediate sum device 1111c may receive status information regarding the number of operable storage devices from the RAID control unit 1111b that can obtain status information from the storage device failure sense unit 1111h.

  The storage device failure sense unit 1111h may determine a state relating to the operation of any storage device coupled to the RAID controller 1111 and determine a list of operational storage devices. In other words, the storage device failure sense unit 1111h may determine whether a given storage device is no longer suitable for reliable storage of data and checksums. The storage device failure sense unit 1111h can test the storage device for reliable operation, can determine whether the given storage device is online, a response from the given storage device is a predetermined timeout If not received within an interval, a unit offline may be declared, a signal quality metric may be determined if the data read from the storage device is less than a threshold quality, an operational storage device may be listed, etc. . The storage device failure sense unit 1111h may record the results of such tests and distribute a list of operational storage devices for elements of the RAID controller 1111 such as the RAID control unit 1111b.

  The intermediate sum calculator 1111d can calculate an intermediate, local and partial sum, within which the error correction code calculation for the checksum and data is as described with respect to Equation 8 and Equation 13, respectively. Can be broken down into The intermediate or partial sum may be a weighted sum of symbols read from an operational storage device in the cluster that reports to the RAID controller 1111. For example, the corresponding limit on the cluster of storage devices and the sum of these partial sums may be determined from a configuration matrix such as the configuration matrix 200 or other data structure. The intermediate sum calculator 1111d may calculate data and checksum symbols after receiving corresponding partial sums from other RAID controllers such as RAID controller 1112 or RAID controller 1113.

  The recalculator 1111e performs an intermediate local checksum based on data from the storage device directly coupled to the RAID controller 1111 and a non-local intermediate check from other RAID controllers transferred through the communication fabric interface 1111a. Sam can be recalculated. In other words, if the data or checksum symbol change occurs in a local storage device that is directly coupled to the RAID controller 1111 or is caused by an intermediate checksum sent to the RAID controller 1111 through the communication fabric 1200, the recalculation Unit 1111e may modify the result from intermediate sum calculator 1111d accordingly.

  The calculation control 1111f controls both the intermediate sum calculator 1111d and the recalculator 1111e to determine whether the intermediate checksum calculation result or the recalculated checksum should be used for FEC. sell. The RAID control 1111b can signal the calculation control 1111f directly or through the communication fabric interface 1111a to determine whether the result of the intermediate sum calculator 1111d or the result of the recalculator 1111e is calculated. . The RAID control 1111b can obtain state information such as a state relating to an operation related to the data storage device from the storage device failure sense unit 1111h.

によって、チェックサムを計算し記憶しうる。式中、インデックスｔは、１からＲＡＩＤコントローラの数ｒまでの範囲にあることができ、ｃ_ｉ，ｔは、ｔ番目のインデックスについてのｉ番目の中間チェックサムである。たとえば、ｒは、並列ＲＡＩＤアーキテクチャ１００の場合、３に等しい。構成行列２００に関して述べたように、ＱＳ（ｔ）およびＱＥ（ｔ）は、開始および終了記憶デバイスをＲＡＩＤコントローラにマッピングし、それぞれの中間チェックサムｃ_ｉ，ｊを生成する部分和の合計の制限を確定しうる。関数Ｊ（・）は、たとえば、Ｊ（１００２）＝２であるように、オフセットを減算しうる。 The parallel RAID controllers 1111 to 1113 are

Can calculate and store the checksum. Where index t can range from 1 to the number r of RAID controllers, and c _{i, t} is the i th intermediate checksum for the t th index. For example, r is equal to 3 for the parallel RAID architecture 100. As described with respect to the configuration matrix 200, QS (t) and QE (t) limit the sum of the partial sums that map the start and end storage devices to the RAID controller and generate the respective intermediate checksums c _{i, j.} Can be determined. The function J (•) may subtract the offset such that J (1002) = 2, for example.

によって、中間チェックサムｃ_ｉ，ｔを計算しうる。 The t-th RS-RAID controller such as the RAID controller 1111 is

_Can calculate an intermediate checksum c _{i, t} .

The use of intermediate checksums c _{i, t} can reduce data traffic on communication fabric 1200 and can increase the throughput of parallel RS-RAID architecture 100. For example, in an 8 + 4 RS-RAID architecture, if a single primary RAID controller controls all of the storage devices and calculates the checksum, eight data symbols can be transferred through the communication fabric. In contrast, using the intermediate checksum calculator results from the 8 + 4 parallel RS-RAID architecture, only two intermediate checksum symbols may need to be transferred through the communication fabric.

によって、データ差に重み付けしうる。 In addition to calculating the intermediate checksum and the full checksum, the parallel RS-RAID architecture 100 may modify or maintain the checksum symbol as the data symbol changes. For example, if the data symbol changes from d _j to d ′ _j , the checksum is
c ′ _i = c _i + f _{i, j} (d ′ _j −d _j ) Equation 9
Can be recalculated. When performing the calculation of Equation 9, the RAID controller 1111 calculates the data difference (d ′ _j −d _j ), and the van der Monde element fi _{, j} ,

Can weight the data difference.

Individual parallel RAID controllers 1111 to 1113 can send temporary components c ′ _{i, t} to other controllers of RAID controllers 1111 to 1113.

によって、それぞれの割当てられた記憶デバイスを更新しうる。 The RS-RAID controllers 1111 to 1113 are

Can update each assigned storage device.

  When a storage device fails, for example, when the storage device failure sense unit 1111h detects a hard disk crash, the inverse matrix Inv (A) of the expanded matrix is modified by the parallel RAID controllers 1111-1113 to leave the remaining or ready data An inverse matrix Inv (A ′) corresponding to the storage device is formed. The matrix Inv (A ′) may be a static data structure unless further storage device failure occurs. If another storage device fails, Inv (A ′) can be calculated once and then broadcast to all operational RAID controllers, such as RAID controllers 1111 to 1113. If more storage devices fail later, a new inverse matrix Inv (A ″) may be recalculated and broadcast to all RAID controllers.

は、

から回復される可能性がある。式中、Ｉｎｖ（Ａ'）の要素は、ａ_ｉ，ｊ（１≦ｉ≦ｎおよび１≦ｊ≦ｎ）で示されてもよい。対応するデータおよびチェックサムシンボルＥ'＝［ｅ_１、ｅ_２、…、ｅ_ｎ］^Ｔの要素は、作動可能でかつ選択されたデータ記憶デバイスから読出されうる。並列ＲＳ−ＲＡＩＤアーキテクチャ１００は、Ｅの行ならびに添加された単位行列とファンデルモンデ行列の対応する部分集合を選択するかまたは選抜して、それぞれ、Ｅ'およびＩｎｖ（Ａ'）が形成されうる。換言すれば、並列ＲＳ−ＲＡＩＤアーキテクチャ１００は、データ回復計算を、

によって、部分和または中間データシンボルの集合に分解しうる。式中、ｅ_ｊは、ｔ番目のＲＳ−ＲＡＩＤコントローラの制御下にある全てのデータまたはチェックサムシンボルの集合であると理解される。 The parallel RS-RAID architecture 100 can recover data symbols using intermediate sums or partial sums calculated locally at each RAID controller even if a storage device fails. Data recovered

Is

May be recovered from. In the formula, the element of Inv (A ′) may be represented by a _{i, j} (1 ≦ i ≦ n and 1 ≦ j ≦ n). Corresponding data and elements of the checksum symbol E ′ = [e ₁ , e ₂ ,..., E _n ] ^T are operational and can be read from the selected data storage device. Parallel RS-RAID architecture 100 may select or select E rows and corresponding subsets of added identity and van der Monde matrices to form E ′ and Inv (A ′), respectively. . In other words, the parallel RS-RAID architecture 100 performs data recovery computation as follows:

Can be decomposed into a set of partial sums or intermediate data symbols. _Where ej is understood to be the set of all data or checksum symbols under the control of the t th RS-RAID controller.

を計算し、次に、

によって、回復されるデータ

を計算しうる。 When a message, such as an intermediate data symbol, is received from another parallel RS-RAID controller, each RAID controller first starts with an intermediate data symbol

And then

Recovered data by

Can be calculated.

  FIG. 4A shows an example of a checksum program flowchart 400A for a parallel RS-RAID architecture for data storage. Program flowchart 400A may begin at program step S410 and proceed to program step S420, where a parallel RS-RAID architecture configuration matrix may be read. For example, the configuration matrix may specify starting and ending device numbers for storage devices associated with a given RAID controller, such as the configuration matrix described with respect to FIG. Each RAID controller may store a local copy of the configuration matrix, may reconcile the configuration matrix with other RAID controllers, may receive the configuration matrix from a higher level RAID device or network controller, etc. May be understood.

  From program step S420, the program flow can proceed to program step S425, where the program can read the stored information holding data symbols from the external device. For example, the program may receive a set of 2K bit data blocks received from the flash drive through the communication fabric.

From program step S425, the program can proceed to program step S430 where an intermediate checksum can be calculated. For example, an intermediate checksum or data and parity calculation may be calculated from a linear combination of data words using Equation 8 and Galois field properties. Program step S430 calculates an intermediate checksum and 1) uses the stored codeword symbols from each operable storage unit to supply a given portion of the codeword symbols to a given RAID controller. The sum may be updated or maintained, and 2) intermediate checksums from other RAID controllers communicating with a given RAID controller through the communication fabric may be aggregated. In other words, low latency data and parity bits, bytes, or words from the local subset data storage unit are stored in high latency, in the form of ci _{, t} , from other data storage units. Or may be combined with partially summed data and parity. Thereafter, the program proceeds to program step S460.

In step S460, the program may distribute the intermediate checksum to different RS-RAID controllers. For example, if Q (t) = t, program step S460 may distribute the first intermediate checksum c _1,1 from the first RAID controller to the second and third RAID controllers.

From program step S460, program flow can proceed to program step S470, where the program can receive an intermediate checksum from another RAID controller. From program step S470, the program can proceed to program step S480. The set of intermediate checksums allows each RAID controller to calculate a complete checksum c _{i according} to Equation 8 and store c _i for subsequent error correction and detection calculations. For example, the program can receive the second and third intermediate checksum _{c i, 2,} and _{c i, 3,} second and third intermediate checksum _{c i, 2,} and _{c i, 3} is With the first checksum c _{i, 1} calculated locally, a sufficient set of checksums to calculate c ₁ may be formed.

  From program step S480, program flow can proceed to program step S490, where the program can store data assigned to the RAID controller executing the program and a complete checksum symbol. For example, a program may strip data and checksum symbols across an array of disks. From program step S490, the program flow can proceed to program step S495 where program execution can stop.

  FIG. 4B shows an example of a checksum update program flowchart 400B for a parallel RS-RAID architecture for data storage. Program flowchart 400B starts in step S440 and proceeds to step S442.

  At step S442, the parallel RS-RAID architecture may receive a data change. For example, the storage device may receive new data symbols to replace old data symbols. Thereafter, the program flow can proceed to step S444.

  In step S444, the RAID controller coupled to the storage device may calculate a temporary component according to Equation 10. The RAID controller may obtain the data difference between the new data symbol and the old data symbol and weight the data difference by van der Monde matrix elements. Thereafter, the program flow can proceed to step S446.

  In step S446, the temporary component can be communicated to other RAID controllers. In certain embodiments, the communication fabric may combine various RAID controllers. The communication fabric may communicate a temporary component corresponding to the data change to a RAID controller that controls the storage device that stores the checksum. Thereafter, the program flow may proceed to step S448.

  In step S448, the RAID controller that controls the storage device that stores the checksum may update the checksum based on the received temporary component, for example, using Equation 11. Thereafter, the program flow may proceed to step S450 and stop.

  FIG. 5 shows an example of a data program flow chart 500 for a parallel RS-RAID architecture for data storage. Program flowchart 500 may begin at step S510 and proceed to step S520, where a configuration matrix for a parallel RS-RAID architecture may be read as described with respect to FIG. 4A. From program step S520, program flow can proceed to program step S525, where data and checksum symbols can be read from the storage device. For example, 8 data and 4 checksum symbols can be read from 12 storage devices. In this example, at least eight data or checksum symbols can be read from the operable storage device.

From program step S425, the program flow can proceed to program step S530, where the program can calculate intermediate data symbols. For example, the program may calculate intermediate data symbols according to Equation 13. The weighting factors a _{i, j} used in Equation 13 may be calculated in advance and distributed to the RAID controller, or may be recalculated as necessary, such as after reading the configuration matrix in program step S520. It may be understood that it is good. From program step S530, program flow can proceed to program step S540, where the program can distribute the intermediate data symbols to the parallel RAID controller.

  From program step S540, the program flow can proceed to program step S550, where the program can receive intermediate data symbols from the parallel RAID controller. From program step S550, program flow can proceed to program step S560, where the program recovers data symbols from intermediate data symbols that came from both the local RAID controller and the parallel RAID controller. Can be calculated. In other words, the program may sum the intermediate data symbols according to Equation 14. From program step S560, the program flow can proceed to program step S570 where program execution can stop.

  Although the invention has been described with reference to specific exemplary embodiments of the invention, it will be apparent that many alternatives, modifications, and variations will be apparent to those skilled in the art. Accordingly, the embodiments of the invention described herein are intended to be illustrative rather than limiting. There are changes that may be made without departing from the spirit and scope of the invention.

Claims (26)

A distributed data storage device,
A plurality of parallel data storage clusters,
A plurality of data storage devices for storing a plurality of data symbols; and
Coupling to each of the plurality of data storage devices, calculating a local intermediate checksum from the plurality of data symbols, and checking the plurality of data symbols based on a plurality of intermediate checksums including the local intermediate checksum A plurality of parallel data storage clusters including a RAID controller configured to calculate a sum;
A distributed data storage device comprising: a communication fabric coupled to each of the plurality of parallel data storage clusters and delivering the plurality of intermediate checksums to the plurality of parallel data storage clusters. The distributed data storage device of claim 1, wherein the RAID controller calculates the local intermediate checksum based on a weighted sum of the plurality of data symbols. At least one of the plurality of RAID controllers of the plurality of parallel data storage clusters receives the intermediate checksum from another parallel data storage cluster, and is configured to include a plurality of data storage devices of the plurality of parallel data storage clusters. The distributed data storage device according to claim 1 or 2, wherein a checksum for storing in one is generated based on the received intermediate checksum and the local intermediate checksum. The RAID controller, distributed data storage device according to any one of the three further from configured claim 1 to store the checksum in the data storage device. The communication fabric, distributed data storage device according to the plurality of information symbols from further configured claim 1 to deliver to the plurality of parallel data storage clusters to any one of 4, including a plurality of data symbols. 6. The distributed data storage device of claim 5 , wherein the RAID controller further comprises a RAID control unit that selects the plurality of data symbols from the subset of the plurality of information symbols. The distributed data storage device of claim 6 , wherein the RAID control unit further comprises a storage device failure sense unit that establishes a list of operable data storage devices from the plurality of data storage devices. The distributed data storage device of claim 7 , wherein the RAID control unit is further configured to calculate a data recovery matrix based on the list of operable data storage devices. The distributed data storage device of claim 8 , wherein the communication fabric is further configured to deliver the data recovery matrix to the plurality of parallel data storage clusters when the list of operational data storage devices changes. The distributed data storage device of claim 6 , wherein the RAID controller further comprises an intermediate sum device configured to calculate the intermediate checksum based on a list of operational data storage devices. The distributed data storage device of claim 10 , wherein the intermediate sum device further comprises a recalculator configured to update the local intermediate checksum when a data symbol changes. 9. The distributed data storage device of claim 8 , wherein the RAID control unit is further configured to determine an inverse matrix of an added unit matrix and van der Monde matrix to calculate the data recovery matrix. Intermediate sum device intermediate data symbols based on the data recovery matrix, and, according to further configured claim 12 to calculate at least one of the vectors of the read data symbols and the read checksum symbol is read from the storage device Distributed data storage device. The RAID control unit selects a vector based on the set of rows of the added identity matrix and van der Monde matrix and the list of operable data storage devices to form the data recovery matrix. The distributed data storage device of claim 12 , further configured to determine an inverse matrix of the set. 15. The distributed data storage device of claim 14 , wherein the RAID controller is further configured to send a message to the plurality of parallel data storage clusters that causes an intermediate sum device to calculate intermediate data symbols. The distributed data storage device of claim 15 , wherein the communication fabric is further configured to distribute the intermediate data symbols to each of the plurality of parallel data storage clusters. The distributed data storage device of claim 16 , wherein the RAID controller is further configured to calculate recovered data symbols from a plurality of intermediate data symbols. Said local intermediate checksum, any one of the partial sum of a subset of said plurality of data symbols corresponding to said plurality of data storage devices assigned to the RAID controller (partials sum) at a claims 1 to 17 The distributed data storage device according to item . 14. The distributed data storage device of claim 13 , wherein the intermediate data symbols are a subset of the data symbols corresponding to the data storage device assigned to the RAID controller and a partial sum of the read checksum symbols. A data storage method,
Assigning multiple data storage devices to multiple parallel data storage clusters;
Storing a plurality of data symbols in each of the plurality of parallel data storage clusters;
Each of a plurality of RAID controllers included in each of the plurality of parallel data storage clusters calculates an intermediate checksum from a weighted sum of the plurality of data symbols;
At least one of the plurality of RAID controllers receives the intermediate checksum from each of the other RAID controllers of the plurality of RAID controllers ;
At least one of the plurality of RAID controllers calculates a checksum of the plurality of data symbols based on the calculated intermediate checksum and the intermediate checksum received from each of the other RAID controllers ; and
A data storage method comprising storing the checksum in at least one of the plurality of data storage devices. Selecting the plurality of data symbols from a subset of information symbols; and
21. The data storage method of claim 20 , further comprising distributing the selected data symbols to the plurality of parallel data storage clusters using a communication fabric. The data storage method of claim 21 , further comprising detecting a set of operable data storage devices in each of the plurality of parallel data storage clusters. 23. The data storage method of claim 22 , further comprising distributing a data recovery matrix based on the set of operable data storage devices to each of the plurality of parallel data storage clusters. 24. The data storage method of claim 23 , further comprising calculating a vector of at least one of an intermediate data symbol based on the data recovery matrix and a read data symbol and a read checksum symbol read from the storage device. Delivering the intermediate data symbols to each of the plurality of parallel data storage clusters; and
The data storage method of claim 24 , further comprising calculating a recovered data symbol from the sum of the intermediate data symbols. A method for error correction in a distributed data architecture, comprising:
Reading a configuration matrix that allocates a plurality of data storage devices to a plurality of parallel data storage clusters;
Storing a plurality of data symbols in the plurality of parallel data storage clusters;
Each of a plurality of RAID controllers included in each of the plurality of parallel data storage clusters calculates an intermediate checksum from the plurality of data symbols stored in the assigned data storage device, thereby Calculating at least one intermediate checksum for each data storage cluster;
At least one of the plurality of RAID controllers receives the intermediate checksum from each of the other RAID controllers of the plurality of RAID controllers ;
At least one of the plurality of RAID controllers sums the calculated intermediate checksum and the intermediate checksum received from each of the other RAID controllers to form a checksum; and
A method of storing the checksum in at least one data storage device.

Images